ngle_accel.c revision 852b373f 
1/*
2 * NGLE - hardware acceleration.
3 *
4 * Copyright (C) 2024 Michael Lorenz
5 *
6 * Permission is hereby granted, free of charge, to any person obtaining a copy
7 * of this software and associated documentation files (the "Software"), to deal
8 * in the Software without restriction, including without limitation the rights
9 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
10 * copies of the Software, and to permit persons to whom the Software is
11 * furnished to do so, subject to the following conditions:
12 *
13 * The above copyright notice and this permission notice shall be included in
14 * all copies or substantial portions of the Software.
15 *
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
19 * MICHAEL LORENZ BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER
20 * IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
21 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
22 */
23
24/* $NetBSD: ngle_accel.c,v 1.5 2024/10/25 09:20:37 macallan Exp $ */
25
26#include <sys/types.h>
27#include <dev/ic/stireg.h>
28
29
30#include "ngle.h"
31
32//#define DEBUG
33
34#ifdef DEBUG
35#define ENTER xf86Msg(X_ERROR, "%s\n", __func__)
36#define LEAVE xf86Msg(X_ERROR, "%s done\n", __func__)
37#define DBGMSG xf86Msg
38#else
39#define ENTER
40#define DBGMSG if (0) xf86Msg
41#define LEAVE
42#endif
43
44static inline void
45NGLEWrite4(NGLEPtr fPtr, int offset, uint32_t val)
46{
47	volatile uint32_t *ptr = (uint32_t *)((uint8_t *)fPtr->regs + offset);
48	*ptr = val;
49}
50
51static inline void
52NGLEWrite1(NGLEPtr fPtr, int offset, uint8_t val)
53{
54	volatile uint8_t *ptr = (uint8_t *)fPtr->regs + offset;
55	*ptr = val;
56}
57
58static inline uint32_t
59NGLERead4(NGLEPtr fPtr, int offset)
60{
61	volatile uint32_t *ptr = (uint32_t *)((uint8_t *)fPtr->regs + offset);
62	return *ptr;
63}
64
65static inline uint8_t
66NGLERead1(NGLEPtr fPtr, int offset)
67{
68	volatile uint8_t *ptr = (uint8_t *)fPtr->regs + offset;
69	return *ptr;
70}
71
72static void
73NGLEWaitMarker(ScreenPtr pScreen, int Marker)
74{
75	ScrnInfoPtr pScrn = xf86Screens[pScreen->myNum];
76	NGLEPtr fPtr = NGLEPTR(pScrn);
77	uint8_t stat;
78
79	ENTER;
80	do {
81		stat = NGLERead1(fPtr, NGLE_REG_15b0);
82		if (stat == 0)
83			stat = NGLERead1(fPtr, NGLE_REG_15b0);
84	} while (stat != 0);
85	LEAVE;
86}
87
88static void
89NGLEWaitFifo(NGLEPtr fPtr, int slots)
90{
91	uint32_t reg;
92
93	ENTER;
94	do {
95		reg = NGLERead4(fPtr, NGLE_REG_34);
96	} while (reg < slots);
97	LEAVE;
98}
99
100static Bool
101NGLEPrepareCopy_EG
102(
103    PixmapPtr pSrcPixmap,
104    PixmapPtr pDstPixmap,
105    int       xdir,
106    int       ydir,
107    int       alu,
108    Pixel     planemask
109)
110{
111	ScrnInfoPtr pScrn = xf86Screens[pDstPixmap->drawable.pScreen->myNum];
112	NGLEPtr fPtr = NGLEPTR(pScrn);
113	int srcpitch = exaGetPixmapPitch(pSrcPixmap);
114	int srcoff = exaGetPixmapOffset(pSrcPixmap);
115
116	ENTER;
117
118	DBGMSG(X_ERROR, "%s %d %d\n", __func__, srcoff, srcpitch);
119	fPtr->offset = srcoff >> 11;
120	NGLEWaitMarker(pDstPixmap->drawable.pScreen, 0);
121	NGLEWrite4(fPtr, NGLE_REG_10,
122	    BA(IndexedDcd, Otc04, Ots08, AddrLong, 0, BINapp0I, 0));
123	NGLEWrite4(fPtr, NGLE_REG_14,
124	    IBOvals(RopSrc, 0, BitmapExtent08, 1, DataDynamic, MaskOtc, 0, 0));
125	NGLEWrite4(fPtr, NGLE_REG_13, planemask);
126
127	fPtr->hwmode = HW_BLIT;
128
129	LEAVE;
130	return TRUE;
131}
132
133static Bool
134NGLEPrepareCopy_HCRX
135(
136    PixmapPtr pSrcPixmap,
137    PixmapPtr pDstPixmap,
138    int       xdir,
139    int       ydir,
140    int       alu,
141    Pixel     planemask
142)
143{
144	ScrnInfoPtr pScrn = xf86Screens[pDstPixmap->drawable.pScreen->myNum];
145	NGLEPtr fPtr = NGLEPTR(pScrn);
146	int srcpitch = exaGetPixmapPitch(pSrcPixmap);
147	int srcoff = exaGetPixmapOffset(pSrcPixmap);
148
149	ENTER;
150
151	DBGMSG(X_ERROR, "%s %d %d\n", __func__, srcoff, srcpitch);
152	fPtr->offset = srcoff / srcpitch;
153	NGLEWaitMarker(pDstPixmap->drawable.pScreen, 0);
154	NGLEWrite4(fPtr, NGLE_REG_10,
155	    BA(FractDcd, Otc24, Ots08, AddrLong, 0, BINapp0F8, 0));
156	NGLEWrite4(fPtr, NGLE_REG_14,
157	    IBOvals(RopSrc, 0, BitmapExtent32, 0, DataDynamic, MaskOtc, 0, 0));
158	NGLEWrite4(fPtr, NGLE_REG_13, planemask);
159
160	fPtr->hwmode = HW_BLIT;
161
162	LEAVE;
163	return TRUE;
164}
165
166static void
167NGLECopy
168(
169    PixmapPtr pDstPixmap,
170    int       xs,
171    int       ys,
172    int       xd,
173    int       yd,
174    int       wi,
175    int       he
176)
177{
178	ScrnInfoPtr pScrn = xf86Screens[pDstPixmap->drawable.pScreen->myNum];
179	NGLEPtr fPtr = NGLEPTR(pScrn);
180	int dstpitch = exaGetPixmapPitch(pDstPixmap);
181	int dstoff = exaGetPixmapOffset(pDstPixmap);
182
183	ENTER;
184	NGLEWaitFifo(fPtr, 3);
185	NGLEWrite4(fPtr, NGLE_REG_24, (xs << 16) | (ys + fPtr->offset));
186	NGLEWrite4(fPtr, NGLE_REG_7, (wi << 16) | he);
187	NGLEWrite4(fPtr, NGLE_REG_25, (xd << 16) | (yd + (dstoff / dstpitch)));
188
189	exaMarkSync(pDstPixmap->drawable.pScreen);
190	LEAVE;
191}
192
193static void
194NGLEDoneCopy(PixmapPtr pDstPixmap)
195{
196    ENTER;
197    LEAVE;
198}
199
200static Bool
201NGLEPrepareSolid_EG(
202    PixmapPtr pPixmap,
203    int alu,
204    Pixel planemask,
205    Pixel fg)
206{
207	ScrnInfoPtr pScrn = xf86Screens[pPixmap->drawable.pScreen->myNum];
208	NGLEPtr fPtr = NGLEPTR(pScrn);
209
210	ENTER;
211	NGLEWaitFifo(fPtr, 4);
212	/* plane mask */
213	NGLEWrite4(fPtr, NGLE_REG_13, planemask);
214	/* bitmap op */
215	NGLEWrite4(fPtr, NGLE_REG_14,
216	    IBOvals(alu, 0, BitmapExtent08, 0, DataDynamic, MaskOtc, 1, 0));
217	/* dst bitmap access */
218	NGLEWrite4(fPtr, NGLE_REG_11,
219	    BA(IndexedDcd, Otc32, OtsIndirect, AddrLong, 0, BINapp0I, 0));
220    	NGLEWrite4(fPtr, NGLE_REG_35, fg);
221	fPtr->hwmode = HW_FILL;
222
223	LEAVE;
224	return TRUE;
225}
226
227static Bool
228NGLEPrepareSolid_HCRX(
229    PixmapPtr pPixmap,
230    int alu,
231    Pixel planemask,
232    Pixel fg)
233{
234	ScrnInfoPtr pScrn = xf86Screens[pPixmap->drawable.pScreen->myNum];
235	NGLEPtr fPtr = NGLEPTR(pScrn);
236
237	ENTER;
238	NGLEWaitFifo(fPtr, 4);
239	/* plane mask */
240	NGLEWrite4(fPtr, NGLE_REG_13, planemask);
241	/* bitmap op */
242	NGLEWrite4(fPtr, NGLE_REG_14,
243	    IBOvals(alu, 0, BitmapExtent32, 0, DataDynamic, MaskOtc, 1, 0));
244	/* dst bitmap access */
245	NGLEWrite4(fPtr, NGLE_REG_11,
246	    BA(FractDcd, Otc32, OtsIndirect, AddrLong, 0, BINapp0F8, 0));
247    	NGLEWrite4(fPtr, NGLE_REG_35, fg);
248	fPtr->hwmode = HW_FILL;
249
250	LEAVE;
251	return TRUE;
252}
253
254static void
255NGLESolid(
256    PixmapPtr pPixmap,
257    int x1,
258    int y1,
259    int x2,
260    int y2)
261{
262	ScrnInfoPtr pScrn = xf86Screens[pPixmap->drawable.pScreen->myNum];
263	NGLEPtr fPtr = NGLEPTR(pScrn);
264	int w = x2 - x1, h = y2 - y1;
265	int pitch = exaGetPixmapPitch(pPixmap);
266	int offset = exaGetPixmapOffset(pPixmap);
267	uint32_t mask;
268	int wi, rest;
269
270	ENTER;
271
272	y1 += offset / pitch;
273
274	/*
275	 * XXX
276	 * Turns out this thing always fills rectangles to the next 32 pixel
277	 * boundary on te right. To get around this we split the rectangle
278	 * into a multiples-of-32 part and the rest, so we can mask off the
279	 * excess pixels.
280	 */
281	rest = w & 0x1f;
282	wi = w & 0xffffe0;
283	if (wi > 0) {
284		NGLEWaitFifo(fPtr, 3);
285		/* transfer data */
286		NGLEWrite4(fPtr, NGLE_REG_8, 0xffffffff);
287		/* dst XY */
288		NGLEWrite4(fPtr, NGLE_REG_6, (x1 << 16) | y1);
289		/* len XY start */
290		NGLEWrite4(fPtr, NGLE_REG_9, (wi << 16) | h);
291	}
292	if (rest > 0) {
293		mask = 0xffffffff << (32 - w);
294		/* transfer data */
295		NGLEWaitFifo(fPtr, 3);
296		NGLEWrite4(fPtr, NGLE_REG_8, mask);
297		/* dst XY */
298		NGLEWrite4(fPtr, NGLE_REG_6, ((x1 + wi) << 16) | y1);
299		/* len XY start */
300		NGLEWrite4(fPtr, NGLE_REG_9, (rest << 16) | h);
301	}
302	exaMarkSync(pPixmap->drawable.pScreen);
303	LEAVE;
304}
305
306Bool
307NGLEPrepareAccess_EG(PixmapPtr pPixmap, int index)
308{
309	ScrnInfoPtr pScrn = xf86Screens[pPixmap->drawable.pScreen->myNum];
310	NGLEPtr fPtr = NGLEPTR(pScrn);
311
312	if (fPtr->hwmode == HW_FB) return TRUE;
313
314	NGLEWaitMarker(pPixmap->drawable.pScreen, 0);
315	NGLEWrite4(fPtr, NGLE_REG_10,
316	    BA(IndexedDcd, Otc04, Ots08, AddrByte, 0, BINapp0I, 0));
317	NGLEWrite4(fPtr, NGLE_REG_14, 0x83000300);
318	NGLEWrite4(fPtr, NGLE_REG_13, 0xff);
319	NGLEWaitMarker(pPixmap->drawable.pScreen, 0);
320	NGLEWrite1(fPtr, NGLE_REG_16b1, 1);
321	fPtr->hwmode = HW_FB;
322	return TRUE;
323}
324
325Bool
326NGLEPrepareAccess_HCRX(PixmapPtr pPixmap, int index)
327{
328	ScrnInfoPtr pScrn = xf86Screens[pPixmap->drawable.pScreen->myNum];
329	NGLEPtr fPtr = NGLEPTR(pScrn);
330
331	if (fPtr->hwmode == HW_FB) return TRUE;
332
333	NGLEWaitMarker(pPixmap->drawable.pScreen, 0);
334	NGLEWrite4(fPtr, NGLE_REG_10,
335	    BA(FractDcd, Otc24, Ots08, AddrLong, 0, BINapp0F8, 0));
336	NGLEWrite4(fPtr, NGLE_REG_14, 0x83000300);
337	NGLEWrite4(fPtr, NGLE_REG_13, 0xffffffff);
338	NGLEWaitMarker(pPixmap->drawable.pScreen, 0);
339	NGLEWrite1(fPtr, NGLE_REG_16b1, 1);
340	fPtr->hwmode = HW_FB;
341	return TRUE;
342}
343Bool
344NGLEInitAccel(ScreenPtr pScreen)
345{
346	ScrnInfoPtr pScrn = xf86Screens[pScreen->myNum];
347	NGLEPtr fPtr = NGLEPTR(pScrn);
348	ExaDriverPtr pExa;
349	int lines, bpp = pScrn->bitsPerPixel >> 3;
350
351	pExa = exaDriverAlloc();
352	if (!pExa)
353		return FALSE;
354
355	fPtr->pExa = pExa;
356
357	pExa->exa_major = EXA_VERSION_MAJOR;
358	pExa->exa_minor = EXA_VERSION_MINOR;
359
360	pExa->memoryBase = fPtr->fbmem;
361	lines = fPtr->fbmem_len / fPtr->fbi.fbi_stride;
362	DBGMSG(X_ERROR, "lines %d\n", lines);
363	pExa->memorySize = fPtr->fbmem_len;
364	pExa->offScreenBase = fPtr->fbi.fbi_stride * fPtr->fbi.fbi_height;
365	pExa->pixmapOffsetAlign = fPtr->fbi.fbi_stride;
366	pExa->pixmapPitchAlign = fPtr->fbi.fbi_stride;
367
368	pExa->flags = EXA_OFFSCREEN_PIXMAPS | EXA_MIXED_PIXMAPS;
369
370	pExa->maxX = 2048;
371	pExa->maxY = 2048;
372
373	fPtr->hwmode = -1;
374
375	pExa->WaitMarker = NGLEWaitMarker;
376	pExa->Solid = NGLESolid;
377	pExa->DoneSolid = NGLEDoneCopy;
378	pExa->Copy = NGLECopy;
379	pExa->DoneCopy = NGLEDoneCopy;
380	switch (fPtr->gid) {
381		case STI_DD_EG:
382			pExa->PrepareCopy = NGLEPrepareCopy_EG;
383			pExa->PrepareSolid = NGLEPrepareSolid_EG;
384			pExa->PrepareAccess = NGLEPrepareAccess_EG;
385			break;
386		case STI_DD_HCRX:
387			pExa->PrepareCopy = NGLEPrepareCopy_HCRX;
388			pExa->PrepareSolid = NGLEPrepareSolid_HCRX;
389			pExa->PrepareAccess = NGLEPrepareAccess_HCRX;
390			break;
391		default:
392			xf86Msg(X_ERROR,
393			    "unsupported device GID %08x\n", fPtr->gid);
394			return FALSE;
395	}
396	NGLEWaitMarker(pScreen, 0);
397
398	return exaDriverInit(pScreen, pExa);
399}
400